A memory feature in ChatGPT

OpenAI has announced a new memory feature in ChatGPT. The idea is that it will ‘remember’ key facts about you to help it respond more effectively. At first glance, this seems like a great idea. The types of examples provided include:

> Remembering your job to make responses more relevant.
> Recalling your preferences for document formatting and applying them by default.
> Retaining details about your family and friends, so if you ask it to do things for them, it will tailor the responses accordingly.

This has the potential to be really useful, eliminating the need to repeat information and providing more relevant information to you. Intuitively, it’s the kind of feature you would want from a personal assistant. It’s only available to a limited number of users, so I can’t comment on its effectiveness.

A few questions to ask:

Will it suffer from the ‘you liked this, so here’s more’ issue prevalent on the internet? We’ve all had this – e.g. I once searched for a spare part for my washing machine. For years afterward, the internet bombarded me with ads for various washing machine parts because it was convinced I was some sort of washing machine parts mega fan. At least there seems to be a feature to delete memories.

What’s happening with the data? It’s certainly building an impressive profile of you, isn’t it? OpenAI states:

“As with any ChatGPT feature, you’re in control of your organization’s data. Memories and any other information on your workspace are excluded from training our models. Users have control over how and when their memories are used in chats. In addition, Enterprise account owners can turn memory off for their organization at any time.”

Cory Doctorow would argue that this is just the beginning, and the path of what happens next is well-trodden. Do look him up if you’re not familiar with his work.

Finally, on a human level, we need to be very cautious about any system that bases it’s actions on our past behaviour. As individuals, we learn and grow. If technology relies too heavily on the past, it might harm personal development and growth.

This may come across as very negative, which isn’t my intention. I think the concept is well worth exploring. I’ve often wished for software that learns how I like to work and becomes more helpful and cooperative, so am looking forward to trying it.